Design of a multi-stage microfluidics system for high-speed flow
cytometry and closed system cell sorting for cytomics
Meggie Grafton1,4,5, Lisa M. Reece2,4,5, Pedro P. Irazoqui1, Byunghoo Jung3, Huw D. Summers6,
Rashid Bashir3,4, James F. Leary1,2,4,5
1Weldon School of Biomedical Engineering, 2Department of Basic Medical Sciences,
3Department of Electrical and Computer Engineering, 4Bindley Bioscience Center,
5Birck Nanotechnology Center, Purdue University, West Lafayette, IN, USA
6Cardiff University, Wales, UK
To produce a large increase in total throughput, a multi-stage microfluidics system (US Patent pending) is being
developed for flow cytometry and closed system cell sorting. The multi-stage system provides for sorting and re-sorting
of cohorts of cells beginning with multiple cells per sorting unit in the initial stages of the microfluidic device and
achieving single cell sorting at subsequent stages. This design theoretically promises increases of 2- or 3-orders of
magnitude in total cell throughput needed for cytomics applications involving gene chip or proteomics analyses of sorted
Briefly, silicon wafers and CAD software were used with SU-8 soft photolithography techniques and used as a mold
to create Y-shaped, multi-stage microfluidic PDMS chips. PDMS microfluidic chips were fabricated and tested using
fluorescent microspheres driven through the chip by a microprocessor-controlled syringe drive and excited on an
inverted Nikon fluorescence microscope. Inter-particle spacings were measured and used as experimental data for
queuing theory models of multi-stage system performance.
A miniaturized electronics system is being developed for a small portable instrument. A variety of LED light sources,
waveguides, and APD detectors are being tested to find optimal combinations for creating an LED-APD configuration at
the entry points of the Y-junctions for the multi-stage optical PDMS microfluidic chips. The LEDs, APDs, and PDMS
chips are being combined into an inexpensive, small portable, closed system sorter suitable for operation inside a
standard biohazard hood for both sterility and closed system cell sorting as an alternative to large, expensive, and
conventional droplet-based cell sorters.
Keywords: microfluidics; flow cytometry; cell sorting; high-throughput; PDMS; LED
1.1 The need for portable, closed microfluidic flow cytometers and cell sorters
Most current microfluidic cytometers are very slow (approximately 100-300 cells/sec) which makes them
unreasonable options for many real biological or biomedical applications due to the quantity of cells of interest,
sometimes present in low frequencies, that needs to be analyzed for meaningful results. Most conventional commercial
cell sorters are currently droplet based, precluding their use for sorting biohazardous material or even live human cells
unless housed in a special BSL-2 or BSL-3 facility. There is a real need for a practical or even reasonably high-speed,
inexpensive, closed-system (nondroplet-based to avoid dangerous aerosols) microfluidic cytometry/cell sorter. This
device needs to be very small and portable: perhaps the size of a PDA, although that longer-term design goal can be
achieved in multiple intermediate size ranges and still be very useful. This reduction in size will allow the new device to
be placed inside a standard biohazard hood. Such a device requires the use of novel instrument design, new photonic
excitation sources, fluorescent thin-film deposition filters, and smaller sensors.
1.2 The need for higher speed microfluidic cytometers/sorters to do “real” applications
We are developing a microfluidic device to perform high-speed microfluidic flow cytometry/cell sorting. A valid
criticism of microfluidic cytometers is that they cannot do serious applications requiring significant numbers of cells
which require at least 5,000 cells/sec and sometimes as many as 100,000 cells/sec. "Ultra high-speed sorting" is a new
realm of cell sorting that requires thinking beyond either droplet-based or conventional fluidic lab-on-a-chip sorting
technologies (Leary, 2005). This will require implementations of massively parallel and possibly "exponential staging"
cell sorting architectures (Leary and Frederickson, 2003). More importantly, these new branching-tree architecture
microfluidic designs allow for continuous enrichment re-sorting of cells from the previous stage to eventual purity in
subsequent stages. We have previously shown that one of the bottlenecks of making intelligent ultra high-speed sorting
Imaging, Manipulation, and Analysis of Biomolecules, Cells, and Tissues VI,
edited by Daniel L. Farkas, Dan V. Nicolau, Robert C. Leif,
Proc. of SPIE Vol. 6859, 68590V, (2008) · 1605-7422/08/$18 · doi: 10.1117/12.764037
Proc. of SPIE Vol. 6859 68590V-1
2008 SPIE Digital Library -- Subscriber Archive Copy
decisions is processing the information (Leary et al, 2001; 2002). Digital signal processing complicates that process by
either providing for complete pulse shape signatures or requiring digital signal processing (DSP) analysis of the
waveform which, while fast, still requires some time. Assuming four stages, cell velocities of 10 cm/sec and switching
times of 10 milliseconds, queuing theory predicts total cell sorting speeds of over 10,000 cells per second. If results from
our proposed branching-tree microfluidic architecture experimentally validate this prediction, sorting speeds from 10,000
cells per second, to an order of magnitude greater than that for particle-based assays, may become possible.
1.3 Concept of an exponentially staging, multi-stage microfluidic cytometer/sorter:
The concept of an "exponentially" scaling multi-stage microfluidic sorter is not to have a single cell in the excitation
source in the upper stages of the sorter, but to try to capture boluses of fluid (analogous to droplets in a droplet sorter)
which have signal-to-noise values indicating probable presence of a rare cell of interest. This is analogous to "enrichment
sorting" whereby there might be several cells in a sorting unit including at least one cell of interest. This bolus of cells is
then passed continuously to the second stage where the inter-cell spacings are increased by narrowing the channel cross
section. This process is continued with subsets of that previous bolus of fluid discarded down alternate channels if they
do not have a S/N ratio indicating probable presence of a cell of interest. We get to single cell sorting at either stage 3 or
stage 4. As described previously (Leary, 1994; Leary, 2000), a sorter for rare cells is more efficient (in terms of yield and
purity) in sorting rare cells by multiple enrichment sorts than by slowing the system down to try to accomplish pure
sorting in a single pass. While the sorting efficiency by this approach will go down as the frequency of cell of interest
goes up, the overall process will work for any sorted cell frequency, especially for those frequencies below 10 percent.
2. MATERIALS AND METHODS
2.1 Development of closed system microfluidic cytometer/sorters:
As described in the background section, development of higher-speed, closed-system microfluidic sorters would
represent a significant advance in the field. Many groups are working on this problem, but we have some different
approaches and significant capabilities in terms of both available personnel and microfabrication capabilities. We do not
propose to use very small microfluidic channels, as these are not very practical for analyzing human cells. Conventional
cell sorters actually have always had a microfluidic flow cell at the heart of the instrument. It is the other elements of the
systems that are large and unwieldy. If those other elements could be miniaturized, the overall system would be small
enough to place in a biohazard hood in order to handle both the input and output of cells from these systems safely and in
a sterile manner. We are using photolithography methods to construct polydimethylsiloxane (PDMS) microchips that are
largely transparent at visible wavelengths. To maintain a closed fluidic environment, we plan to use elastomeric valves
that can be turned on and off with approximately 10 msec switching times. If this is done at a single stage with single
cells, the overall throughput rate of the system is rather slow, approximately 100 cells/sec. But even that rate can be
useful for many applications. But as we have also demonstrated in our pending patent (Leary and Frederickson, 2003), if
the initial stages of a branched tree architecture are not at single cell level, this switching speed can result in total
throughputs in a four-stage system of more than one million cells/sec. Essentially this means that one is sorting, and re-
sorting (allowing coincident cells within sort boluses) continuously in four stages. This is similar to the effects of
"enrichment sorting" whereby cells are sorted at a high rate such that the sorting units contain multiple cells, but at least
one cell of interest. Then the sorted cells are re-run at a slower speed thereby decreasing their coincidence in the sorting
unit. Dr. Leary was one of the original inventors of high-speed enrichment sorting and rare-event analysis methods
(patents: Leary et al, 1993, 1998; Corio and Leary 1993, 1996, 1999; review papers: Leary, 1994; Leary, 2000; Leary,
2005). This microfluidic architecture lends itself naturally to several stages of enrichment sorting all within the same
device on a "single pass". Dr. Leary's present "macro" high-speed sorter does a "data sort" in two stages but only a single
stage physical sort. However, the principle is the same and results in an impressive rate of analysis of more than 150,000
cells/sec that has been successfully demonstrated and used in the "macro" system since 1984 (Cupp, Leary et al, 1984;
Leary et al, patents 1993, 1998; Corio and Leary, patents 1993;1996; 1999; Leary, 1994; Leary et al, 2000; 2001; 2002).
Rare-event cell sampling statistics papers we have previously published (Rosenblatt et al, 1997; Hokanson et al, 1999)
show the sorting speeds necessary to have 95 percent confidence levels for detection of rare cells for minimal residual
disease monitoring (Leary et al, 1999; 2000b) The purity and yield requirements for our applications of high-speed
sorting of cells for gene expression analyses (Szaniszlo et al, 2004) and for screening of combinatorial chemistry
aptamer-bead libraries (Yang et al, 2003; Leary et al, 2005) are similar. With the branching tree architectures, physical
cells (not just data) can be continuously discarded in a multi-stage enrichment sorting which results in extremely high
total sorting rates. We are initially designing simple one- and two-stage microfluidic systems, gradually moving to more
sophisticated, higher speed, multi-stage systems that are designed to be easy to run and operate within a standard
biohazard hood for both BSL-2 level safety and sterility.
Proc. of SPIE Vol. 6859 68590V-2
2.2 Multi-stage, higher speed microfluidic cytometer/sorters:
This section will highlight the methods being used for design, construction and testing of higher speed multi-stage
microfluidic cytometer/sorters. The overall design will be guided by multi-stage and multi-queue models as an extension
of earlier queuing theory models (Leary, 1994; Leary, 2000) we have used for successful design and construction of a
two-stage (data processing) high speed macro-sorter (Leary et al, 2001, 2002; US Patents 1993a, 1993b, 1996, 1998) that
is part of the existing instrumentation being improved in the proposed IDBR. That system had a single fluidic path but a
two-step data processing cytometer/sorter. It first produced total cell throughputs of 100,000 cells/sec single color
fluorescence in 1981 and multi-color fluorescence in 1987. Since speed alone was insufficient to form the complex
sorting decisions required to classify and sort rare cells of frequencies below one cell in a million, a neural network
architecture was proposed and built for the second stage, resulting in three more US patents and multiple publications
( Corio and Leary, 1992, 1996, 1998). The system achieved 15,000 cells/sec rates for 8 raw parameters and 3 computed
parameters that were polynomial functions of the raw parameters. These could serve not only as neural network output
parameters but also multivariate statistical classifier functions, which are similar in structure mathematically but with
different coefficient weighting factors. The system proved capable of sorting rare tumor cells in defined cell mixtures at
high purity down to single cell level. It was used for subsequent single cell DNA typing and sequencing of PTEN tumor
suppressor gene mutations as a demonstration of potential detection of rare circulating tumor cells in minimal residual
disease monitoring of breast cancer (Leary et al, 1998; Leary, 2005).
The resulting high–speed cytometer/sorter was large, complex, expensive, and far from being commercially
manufacturable. But this instrument proved that high-speed flow cytometry and sorting of rare cells was indeed possible
– something that definitely was not considered achievable when that work began. Now neither the feasibility nor the
importance of high-speed flow cytometry and rare-event analysis is questioned. But the design of this complex
instrument, and currently available commercial high-speed sorters, has been driven by the perceived need to sequentially
examine cells one at a time. Generally this has been done through one (or a small number) of expensive lasers as light
sources with 3-15 (or more), relatively large PMT detectors for light signal detection, and using macro volumes
(typically 50 µl to several ml) of cell sample. These instruments typically cost more than half a million dollars and
require considerable expertise to properly operate. Much of this sophistication can be miniaturized, and many parts of its
operation can be automated using internal sensing of experimental conditions and auto-gain capabilities. We will do this
with array sensors that can choose the optimal amplification and configuration for given experiments. While this must
occur after basic feasibility testing of microfluidic sorter prototypes, the importance of designing a simple and largely
self-configuring device, requiring minimal operator expertise, will be kept as an important part of the overall design goal.
Much of microfluidic cytometry has proceeded as if 40 years of flow cytometry and cell sorting, and all of the lessons
(sometimes painfully) learned in those fields, did not exist. This has resulted in very slow and primitive microfluidic
cytometers with, in many cases, microchannels that are too small to achieve total throughputs of more than a few
hundred cells per second. This is one to two orders of magnitude below speeds needed for most of the applications of
flow cytometry, and even worse in terms of rates and purities of cells sorted for genomic and proteomic applications. By
combining flow cytometric and BioMEMS expertise, and taking advantage of new advances in photonic and
microelectronic devices, we feel that we can move the microfluidic cytometry field forward in new and interesting ways.
We should be able to perform the types of experiments that macro flow cytometry and sorting has done successfully for
many years but now in a more cost effective and biosafety conscious manner. It would also open many possibilities for
new cytometry use if we could easily bring the cytometer to the experiment rather than the other way around.
2.3 Proposed general concept design of multi-stage system:
Our overall design challenges both of these existing macro- and microfluidic paradigms. We are building a small,
portable, closed system, fluidic switching, multi-stage, exponentially staging, disposable PDMS microfluidic chip. The
chip will have realistic channel diameters with light and matched detector mini-arrays that can be easily moved to
operate in a standard biosafety hood or out in the field where the cytometer/sorter needs to be taken to the application,
rather than the other way around (Figure 1). The proposed instrument will also be self-configuring in terms of signal
processing amplifier gain and sorting delays. Automatic choice of proper amplification, and the ability to accommodate
wide dynamic ranges of signals, will be accomplished through the use of APD arrays that will apply different post-APD
gains from which we can choose the best combinations. It should also be appreciated that the microfluidic chip is
actually a dynamically configurable microfluidic and signal processing array that can be optimally configured in
software for each specific application. The proposed instrument will also be self-configuring in terms of signal
processing amplifier gain and sorting delays. Automatic choice of proper amplification, and the ability to accommodate
wide dynamic ranges of signals, will be accomplished through the use of APD arrays that will apply different post-APD
gains from which we can choose the best combinations. It should also be appreciated that the microfluidic chip is
Proc. of SPIE Vol. 6859 68590V-3
Parallel/Exponential Microfluidics for
Microfluidic Sorting using Elastomeric Valves
Air-controlled Elantomeric Valven to open/clone channels
o One cell type
• A different cell type
LED = superluminescent light emitting diode excitation source
AFD = avalanche photodiode detector
Signal processing electronics and sort control logic switch
The two valves must act together,
one open and the other closed.
Creation of a 3D sandwich system of a disposable PDMS
microfluidic chip between an excitation array and a
I I !
Figure 2: Conceptual schematic of a high-speed, multi-stage,
exponentially staging microfluidic cytometer/sorter. By a
combination of decreasing the channel cross sections and
removing boluses of fluid containing cells not-of-inters, desired
cells can be sorted and re-sorted to ever-increasing purity.
Figure 3: Microfluidic valve sorting requires
that channels be opened and closed in
asynchrony so that the fluid always has a path
open when a given bolus of fluid containing one
or more cells is being diverted for cell sorting.
actually a dynamically configurable microfluidic and signal processing array that can be optimally configured in
software for each specific application.
2.4 Design of a high-speed, multistage, microfluidic sorter:
The multi-stage system (Figure 2) is not necessarily single cell analysis in its initial stages. The principle is to sort
boluses of fluid containing potential cells of interest, if it leads to a useful signal-to-noise (S/N) increase (e.g. S/N > 2).
As an example, if we assume that a fluorescent positive labeled cell has a single cell S/N ratio of 100 times background
cells, then a single fluorescent cell present at a rate of approximately 1 per 100 nearby negative cells will have an overall
group S/N ratio of 2:1. By comparison, a bolus of fluid containing all negative cells will have a S/N ratio of 1. So a bolus
of fluid having a S/N ratio of > 2 will likely contain at least one positive cell of interest amidst all the surrounding
negative cells. That bolus is then sorted down one pathway while boluses of fluid containing purely negative cells are
sorted down the other pathway. This depletion strategy gradually concentrates the positive cells stage by stage.
Additionally, one can spread the inter-cell spacings if one narrows the cross sectional areas of successive stage channels.
That increase in inter-cell spacing allows for higher purity sorting at the next sort stage. If one makes these assumptions
and assumes a fluid velocity of 10 cm/sec and a switching time of 10 milliseconds, then the overall throughput rate to
obtain a virtually pure positive cell sample after four stages is approximately 10,000 cells/sec. The system scales roughly
exponentially rather than multiplicatively as it would appear at first glance. This is the basis of a pending US Patent on
high-speed microfluidic sorting (Leary, 2005).
Figure 1: (A) Top view of branched, multistage, microfluidic system for high-speed flow cytometry and cell sorting, (B)
side view of proposed system showing arrays of LEDs on one side of a PDMS microfluidic chip of the general design of
panel A and a corresponding arrays of APD detectors on the other side. The cell inputs and outputs then are configured in
the third dimension such that the PDMS chip can be moved into, or out of, the sandwiched array (Leary et al, 2005, US
Proc. of SPIE Vol. 6859 68590V-4
Concept: "Valve-less" dielectrophoresis based cell
sorting using nanoparticle labeling to selectitvely
increase the dielectric constant of desired cells
(low flow, no
Fluid and Particlea
• Unlabeled human cell
The sorting will be accomplished in one of three ways, depending on the application:
First, PDMS -elastomeric valve sorting systems have achieved switching times of less than 10 milliseconds (Figure 3).
This technique of building valves right into the PDMS microfluidic chips during the microfabrication process of soft
lithography was pioneered many years ago by the Quake lab (Unger et al, 2000). An advantage is that they are cheap to
build, and are part of an integrated, throw-away PDMS chip. Disadvantages are that elastomeric valves require air
sources which in turn require larger auxiliary devices to drive those elastomeric valves. Since they effectively block most
of the fluid flow down a given channel, the design may require that valves on different ports at a branch point be
synchronized so that one is one when the other is closed to minimize the generation of pressure waves. That has to be
done carefully to prevent fluid pressure waves from being generated that might disturb the flow of cells in multistage
systems. The approach does require that alternative pathways be closed and opened in opposition to each other, but that
can be accomplished and the timing does not need to be perfect. Elastomeric sorting has been certainly been successfully
accomplished (Unger et al, 2000) but only for simpler single stage systems. If the valves are opened and closed fairly
quickly, the effects will be rapidly damped. We will measure these perturbations in flow conditions and their effects on
movements of microspheres and cells in the system using measurements of variations in inter-particle spacings.
Second, a new type of magnetoelastic valve has been previously demonstrated (Jackson et al, 2001). PDMS is modified
to contain approximately 50 percent ferric oxide nanoparticles. The channels can then be deformed to block cells from
going down a given channel using electromagnets which can exert sufficient force to close the channel. This is an
attractive approach because switching speeds are much faster and the chip becomes a magnetically controlled chip. No
clumsy external air pumps are needed to run the valves.
Third, a "valve-less" dielectrophoretic sorting approach (Figure 4) can be used which has been demonstrated to work
nicely for bacteria and many microbial cell types (Hu et al, 2005). However, to make this approach work with human
cells, whose dielectric constant is not large enough in isotonic saline solutions (unlike bacterial cells) requires selectively
increasing the dielectric constant. One way to do that is to attach nanoparticles to desired cells using either antibodies or
aptamers. Dr. Leary's laboratory has considerable experience in using nanoparticles to label cell subpopulations of
interest. For example, commercially available quantum dot nanocrystals (Qdots™, Invitrogen, Inc.) not only have usable
fluorescence, but also have dielectric properties to allow Qdot-labeled cells to be dielectrophoretically sorted in a
microfluidic sorter (Figure 4). A greater fraction of nanocrystal fluorescence could be captured in microfluidic systems
using longer in-beam interrogation times to partially compensate for their roughly 10 times longer fluorescence lifetimes.
PDMS microfluidic chips are being designed using Cgraph CAD software (IC Station Tool Suite, Mentor Graphics
Corp., San Jose, CA). Actual channel sizes are measured using the Alpha-Step IQ profilometer (KLA-TENCOR, San
Jose, CA) inside a Class 100 cleanroom facility. Appropriate microfluidic performance is being performed first by direct
visual observation of fluorescent microspheres flowing through the device while on the stage of an inverted fluorescent
microscope then in more detail, as needed, by measurement of inter-particle spacing by use of our high-speed digital
signal processing circuits.
PDMS microfluidic biochips are being constructed using soft photolithography in the cleanrooms of the Birck
Nanotechnology Center at Purdue University. The proximity and ready access to these microfabrication areas allows us
to design single and multi-stage microfluidic channel PDMS chips iteratively based on a combination of theoretical
Figure 4: Concept of dielectrophoretic
sorting of nano-particle labeled human
cells in a microfluidic cytometer/sorter.
Since human cells in isotonic saline,
unlike microbial cells, have too low a
dielectric constant. The dielectric constant
of cells can be selectively increased by
adding antibody or peptide targeted
nanoparticles of the appropriate size and
composition top allow for
dielectrophoretic sorting in a microfluidic
Proc. of SPIE Vol. 6859 68590V-5
modeling, queuing theory, and experimental results. The overall microfabrication process is relatively inexpensive and
fast under these conditions.
2.6 Optical design:
Flow cytometric and sorting systems are by their nature "tightly integrated” in terms of overall engineering design.
The optical design has to be done as part of the total design of the instrument since all of the subcomponents (fluidic,
optical, electronic) are dependent on one another in a fully integrated system. That said, some of the designs we are
exploring for subcomponents of the overall optical portion of the microfluidic flow cytometer/sorter are shown below.
2.6.1 Light excitation sources:
A variety of different light excitation sources are being used during the design and construction of our microfluidic
sorters. The quickest and easiest light source for rapid prototyping is a Nikon Optiphot inverted phase/fluorescence
microscope in our labs with its 100W mercury (Hg) arc lamp and its many Hg excitation frequencies from 355nm to
546nm. This microscope has been extensively modified to have flow cytometry-type optics which allows us to use our
optical filters inter-changeably between our high-speed sorter and the microscope. Our intention is to build a small
portable instrument that can be easily placed inside a standard BSL-2 hood. We will progress from very small, solid state
lasers (still generally too large, although getting smaller and smaller each year!), to very light-weight and inexpensive
superluminescent LEDs (light emitting diodes) which are already very small (and with very low power requirements,
including flashlight batteries or smaller) which are getting more powerful each year (currently >5mW and getting more
powerful, some with built in pre-focusing systems). Clearly arrays of small, inexpensive LEDs of various colors, could
be placed in patterns corresponding to the geometries of the single or multi-stage microfluidic chips. On the horizon are
very small solid state lasers (nearly cell-sized) being developed by one of our UK Optical Biochip project collaborators
(HDS) (Porta and Summers, 2005).
There has likewise been a photonics revolution in terms of photodetectors. While photomultipliers have become
much smaller, more sophisticated, and less power-hungry, we have already started designing systems using avalanche
photo diode (APD) technology. Several years ago when we began this work APDs were quite noisy, and integrated APD
systems were not readily available. Since that time a number of increasingly sophisticated and easier to use systems have
been developed by a variety of vendors (e.g. Hamamatsu, Perkin Elmer, and several others). Recently exciting new APD
systems have been developed by SensL in Cork, Ireland. These new SensL APD systems are attracting a lot of attention
for their innovation and quality and are now distributed by a large number of vendors world-wide. Some of these low-
cost, high-gain (>106) ultra-high quality, 1mm diameter silicon photodiodes are very low power. We will do initial
simpler designs with single APD systems and then progress appropriately to APD arrays. The advantages of APD arrays
are at least two-fold. First, they can be configured to collect different colors of fluorescence at a given interrogation point.
Second, the same color of fluorescence (or light scatter) can be measured at several different gains simultaneously (if the
APD array is appropriately designed). The best mid-range signal can then be automatically chosen, allowing for self
configuration of optimal APD gain. This would greatly simplify operation and training for inexperienced operators.
2.6.3 Optical filters and Thin Film Deposition Filters:
Conventional flow cytometry uses separate optical filters, typically interference-type, that have band-pass
transmission characteristics. We have been using these conventional filters for initial prototyping of microfluidic systems,
both in our modified inverted microscope described previously and also in some of our free-standing microfluidic
systems with LED light sources. But these filters are still "macro" rather than micro filters. While it is certainly possible
to make special small filters or to micro-deposit dyes directly onto optical photodetectors as others have done, and we
will do so as we iteratively design new microfluidic sorters, we will also try an innovative approach of thin–film
deposition filters developed by one of our collaborators, Tim Sands, currently Director of the Birck Nanotechnology
Center. Dr. Sands has previously developed a method for creating optical filters directly onto PDMS microfluidic chips
by thin film deposition (Chediak et al, 2003). We are working to include these techniques into microfluidic optical
biochip devices. If successful, optical filters passing certain wavelengths could be selectively deposited in patterns
directly onto the PDMS chips such that the optical patterns could be changed by changing the inexpensive disposable
PDMS chips rather than needing to change the filters in the device. If successful, specific application PDMS chips could
be placed into a general purpose microfluidic "reader". This would represent a major advance in the potential
commercialization of useful optical biochip technology.
Proc. of SPIE Vol. 6859 68590V-6
Bi0MEMS Sorting Device Electronics
2.7 BioMEMS Sorting Device Electronics:
We give a brief design description in the following subsections of the electronics and software of the data
acquisition and sort control system for our microfluidic cell sorters. The implementation will miniaturize the electronics
and data acquisition of our microfluidic flow cytometer/cell sorter and use DSP (Digital Signal Processing) and FPGA
(Field Programmable Gate Array) electronic chips to incorporate the advances demonstrated by the previously described
high-speed macro-scale sorter. These include both high-speed and single layer neural network architecture for real-time
multivariate statistical classifiers for sophisticated sort decisions. One of us (PPI), is an expert in miniaturization of high-
speed signal processing electronics.
2.7.1 Sorting Electronics Top-Level Design:
A handheld, multi-stage microfluidic cell-sorter will be controlled by a modular mixed-signal electronics design. At
each stage of separation, the bolus will be analyzed and sorted using an identical electronics block interfaced with a
single state-of-the-art Analog Devices ADSP 21369 Sharc digital signal processor (DSP). The DSP is programmed with
real-time firmware combining object oriented C++ code with assembler language subroutines. Decisions made by the
DSP are implemented through the on-board control logic driving the various valves in the MEMS fabricated microfluidic
chambers. Sample analysis is initiated via controlled emission of incoherent narrow-spectrum light from a series of
LEDs. Acquisition of transmitted light through the sample is from tuned APDs. The signal amplitude produced by the
APD is tracked in each case by a peak track-and-hold (PT&H) circuit. The maximum level over a discrete-time interval
is buffered and multiplexed in the analog domain with the output of the other APDs. The multiplexor output is converted
to binary in a 16-bit pipelined analog-to-digital converter (A/D) and fed to the DSP. The PT&H, buffer, multiplexor and
A/D are contained as a single module on an application-specific integrated circuit (ASIC) and can be arrayed to match
the number of sorting stages. This design is shown in Figure 5.
The digital output stream from a set of APDs corresponding to a single stage of separation can be processed in
parallel with the information from all the other stages. The 20 digital application interface (DAI) pins on the ADSP
21369 allow for eight stages of sorting to be processed simultaneously using individual serial ports. Here, we envision
using 4 with capability for growth in future designs. An overview block diagram of the BioMEMS electronics is shown
in Figure 5:
2.7.2 Sub-Component descriptions:
22.214.171.124 Sensors (cf. optics section for more details):
LEDs at various optical frequencies, and matching avalanche photodiodes, available off-the-shelf, will be used for our
application. APDs are calibrated to generate a 1-volt peak amplitude signal in response to illumination of a sample. This
signal is tracked with a unity-gain buffer driving a capacitor through a diode. The capacitor charges up to the peak value
seen in the APD, and this value drives a buffer into the multiplexer. The diode serves to prevent capacitor discharge as a
result of decreasing APD output. Once the buffered peak signal is sampled by the multiplexer, the capacitor is discharged,
and begins tracking the APD anew. The buffers serve to electrically isolate
Figure 5: BioMEMS Sorting
Electronics Block Diagram showing
signal processing from the APD
detectors through the ASICs, then
DSP real-time processing of the
signals before the control logic which
can control the sorting
Proc. of SPIE Vol. 6859 68590V-7
individual component stages from one another. It is possible to digitize each signal individually and then multiplex the
discrete outputs. This approach requires the use of multiple power-hungry A/Ds, one for each channel. By analyzing the
settling time of our on-chip switches, we can multiplex them in the analog domain and control the trigger of the A/D to
ensure high precision across all APDs from a single high-speed A/D. This saves power and lowers space and cost.
126.96.36.199 Signal Acquisition:
We plan to use a 16-bit, 10 kHz/channel pipelined A/D as a compromise or optimal balance between power, space, and
speed in conversion. While this application’s needs are not inherently high-speed, using a high-speed A/D allows us to
acquire all of our samples and then power down for a portion of each cycle, conserving overall power. Additionally it
makes the design more scalable, so increasing the number of stages will not necessitate a new ASIC design. We can
double (or more) the number of stages, by increasing the number of identical circuit modules on the ASIC, increasing the
duty cycle of the A/D, and adding additional parallel ADSP 21369 processors. As an alternative to our pipelined
architecture, all-flash A/Ds are undoubtedly the fastest when it comes to conversion rates. However, for 'n' bit resolution
a flash A/D requires 2(n-1) comparators and is bulky. This becomes unfeasible for higher resolutions. If however the
comparison operation is spread over several clock cycles, the number of comparators required per clock cycle can be
In our design we make, in effect, a 2 stage pipeline. The comparison operation is spread over two clock phases in a
two-stage flash architecture. For an N bit resolution A/D, during the first clock phase the N/2 most significant bits
(MSBs) are resolved. During the second clock phase the resolved MSBs are removed from the input. After this
subtraction, the residue is amplified to the full scale in order to maintain the same dynamic range and re-use reference
voltages, and the remaining N/2 bits are resolved. The approach used in the two-step A/D can be extended further into
the concept of a pipelined A/D, such that several clock phases are used, and only a few bits resolved per stage.
As is with the case of all pipelined architectures, it takes several clock cycles (depending on number of stages) for an
analog value to be digitized to the required resolution. However, after the initial latency period, there is a new digital
output available every clock cycle. The main advantage of using this architecture is its adaptability, allowing the
designer to draw a fine balance between speed, power and resolution. The precision requirements of each pipeline stage
decrease through the pipeline (i.e. the first stage must be most precise, subsequent stages need only be as precise as the
previous stage less the number of bits resolved previously). Thus analog design complexity can be reduced along the
pipeline (less op-amp gain and bandwidth for later stages).
188.8.131.52 Signal Processing:
The ASIC PT&H reset, multiplexor switching, and A/D triggering can be controlled by the Analog Devices ADSP
21369 Digital Signal Processor. This has a 400MHz SIMD SHARC Core, capable of 2.4 GFLOPS (giga Floating point
operations per second) peak performance. This control is accomplished with minimum overhead given the parallel
capabilities of the processor. Our expectation is that we will be making a sorting decision at a rate of 10 kHz from each
of 4 stages simultaneously in order to achieve the desired throughput. To do this in real-time the signal-processing
algorithms must arrive at a decision in less than 60,000 floating-point operations per second (FLOPS). This capability
allows for a significant amount of computational sophistication. If additional processing power is necessary, we can use
multiple parallel processors to each handle a subset of the overall processing. This has the disadvantage of increasing the
overall power budget in a handheld device, but is less important in a portable device placed within a biohazard hood.
Initially we expect to implement our sorting decisions using pre-defined look-up tables mapping specific APD outputs
to individual system states that can be implemented using the microfluidic valve arrays present in the handheld sorter.
This is both a first-pass approach and a backup. The long term approach is to implement real-time decision making
algorithms within the DSP firmware. We have in place a complete operating system combining C++ and assembly-
language algorithms to acquire, process (rudimentary at present), and relay data from a multiplexed digital stream. DSP
output can be displayed directly on an LCD display integrated with the hand-held sorter, or relayed via USB or wireless
to a PC for display and storage.
A PC-based graphical user interface for reprogramming or monitoring DSP function has been developed in Objective-
C. While it is constantly being enhanced, it already allows for device operational verification, operation log download,
and firmware upgrade as necessary.
184.108.40.206 Flow Control:
Microfluidic flow control is accomplished by level-shifted digital outputs from the DAI pins of the DSP to the valves
in the MEMS structure. All pins operate in parallel allowing modification of the instrument state to occur with negligible
overhead cost to the real-time processing capabilities of the device. Level-shifting and appropriate buffering are
Proc. of SPIE Vol. 6859 68590V-8
accomplished with readily available off-the-shelf components. Design of the microfluidics is described previously (refer
to Figure 4).
220.127.116.11 Top-Level Electronics Assembly:
The components of Figure 5: LEDs, APDs, custom ASICs, DSP, fluidic MEMS chips, and off-chip flow-control
hardware are all assembled on a single multi-chip module .In future work, surface-mount packages will be combined
with bare flip-chip bonded dies using existing hardware available to our lab. Assembly is on a custom printed circuit-
board (PCB) fabricated using our CNC milling machine. An optional LCD screen can be mounted on the reverse side of
the PCB module. The entire device will be encased within a durable metal or ceramic case. The electronics will be
hermetically sealed from the fluid input and output ports. Besides these ports, fluids in processing are similarly sealed
from outside exposure. This sealing process will be accomplished using a low-temperature glass sealing technique being
perfected in our lab.
3. DATA AND RESULTS
3.1 Photolithography and SU-8 Mold Fabrication:
A 5”x 5” square mask layout was designed with the CAD-based program ICgraph Basic v2006 1_1.1 (IC Station
Tool Suite, Mentor Graphics Corp., San Jose, CA). Inverted ‘Y’-shaped microfluidics structures were designed in this
program such that the top branch of the structure would be 100 microns wide and that the lower two branches of the ‘Y’
structure would be 50 microns wide. All channels in these structures were designed to be 50 microns in height. 12 ‘Y’
structures with different geometries were designed and incorporated into a single mask layout with the purpose of later
using the mask to construct 12 different PDMS microchips each with different flow properties. A photomask was then
constructed based on this mask design in the Birck Nanotechnology Center.
All photolithography was completed in the Birck Nanotechnology Center Class 100 nanofabrication cleanroom.. Four
inch silicon wafers were thoroughly cleaned by rinsing with deionized water and Piranha solution (a 1:1 (v/v) mixture of
hydrogen peroxide and concentrated sulfuric acid solutions). The wafers were again rinsed with deionized water and
dried with a stream of compressed nitrogen gas and then baked at 120oC for 15 minutes in a hard bake oven. Wafers
were then placed with the polished side facing upwards onto a spinner and SU-8 negative photoresist was slowly poured
onto each wafer to coat the surface. To achieve a uniform SU-8 layer thickness of 50 microns, the wafers were first spun
at 500 rpm for 10 sec with an acceleration of 100 rpm/sec and then at 3,000 rpm for 30 sec with acceleration of 300
rpm/sec. The wafers were soft-baked by heating on a hot plate at 65°C for 3 min and at 95°C for 8 min. The photomask
generated from the submitted mask design was then aligned with the SU-8 coated wafer in a mask aligner (Canon) and
exposed to UV light (350-400 nm) at 23 mJ/cm2-sec for 15 sec. The exposed wafers were heated on a hot plate at 65°C
for 2 min and again at 95°C for 6 min. The microchannel patterns, visible at this point, were developed by placing wafers
in SU-8 developer solution for 3 min, rinsing with isopropyl alcohol, placing in the developer solution again for 3 min,
and rinsing with isopropyl alcohol once more. Wafers were dried using compressed nitrogen gas and hard baked by
heating on a hot plate at 120°C for 5 min to complete the construction of the SU-8 mold.
The Alpha-Step IQ surface profilometer was used to measure the surface dimensions of the completed SU-8 mold.
All measured dimensions were the cross-section diameters. The main branch channel of the branched tree architecture
had measured dimensions of approximately 110µm wide x 40µm high. The two secondary branches had dimensions of
approx. 55 µm wide x 40µm high.
3.2 PDMS Microfluidics Chip Construction:
PDMS microfluidics chips were fabricated outside of the cleanroom. Cured wafers were first surface-coated with
97% trichloro(3,3,3-trifluoropropyl)silane fumes for 30 minutes in a vacuum desiccator. Surface-coated wafers were then
taped, coated side facing upwards, to the bottom of a clean large plastic hinged-lid box using double-sided tape. PDMS
was prepared by weighing out 50g of PDMS base and 5g PDMS curing agent into a weighing dish and stirred for 10
minutes to mix the two components, being careful not to generate contaminants that could make the PDMS optically
impure and/or fragile. Air bubbles were removed from the PDMS solution by covering the weighing dish and allowing it
to sit for 30 minutes undisturbed at room temperature. The de-bubbled PDMS mixture was then slowly poured over the
surface-coated wafer and the wafer was kept at room temperature for 30 min to allow bubbles to dissipate from the
PDMS. The wafer was then covered and left at room temperature for the PDMS to cure overnight.
An incision was made around the PDMS-coated wafer by using a clean razor blade held at a 45° angle. The
polymerized PDMS disk was then carefully removed from the wafer and placed microchannel side down in a clean glass
Petri dish. Individual PDMS chips containing microfluidics channels of various geometries were then cut out using a
razor blade. Portholes for tubing insertion were created using a 22 gauge needle ground to a blunt end open cone and
Proc. of SPIE Vol. 6859 68590V-9
Concept: One Y-stage" branch point showing increase in inter-particle
distances with lowering of branch cross sectional areas even in the absence
of an active sorting step (which would increase the inter-particle distances)
• Cell or particle of
fitted to a 5 ml syringe. Ultramicrobore PTFE Tubing (0.006”ID, 0.016” OD) was cut at a 45° angle and inserted into
the porthole ensuring that the 45° angle was maintained with respect to the direction of flow through the microchannel.
A drop of uncured PDMS solution was used to secure tubing at the insertion site, forming an inlet port. PDMS chips
were cured on hot plate at 70°C for 2 hours and then plasma etched along with clean glass microscope slides to expose
binding sites on both the chips and slides. Etching was performed in a barrel etcher for 30 seconds with oxygen plasma
using 1 Torr vacuum with 200W RF power. Each etched chip was then placed onto separate etched glass slides so that
the microchannels in the chips were facing the etched glass surface of each slide, and then pressed against the glass
slides to ensure adequate binding of the two.
Individual PDMS chips were analyzed for optical clarity by UV-Vis spectrophotometry. Three mm thick PDMS
chips were found to transmit greater than ninety percent of light in the wavelength range 350-800nm (data not shown).
Due to their high transmission (>90 percent) across the spectrum from 350 – 800nm, we can use PDMS for all excitation
and fluorescence wavelengths of current interest. We can also use either a transmitted or epifluorescence
excitation/fluorescence physical design to analyze by fluorescence microscopy.
3.3 Microfluidic Flow Cytometry Application to Test Particle Flow:
To quickly prototype microfluidic PDMS chip designs, we have used mixtures of red and green fluorescent
microspheres: 10µm diameter green fluorescent polystyrene microspheres (Duke Scientific, Palo Alto, CA) and 6µm red
fluorescent AlignFlowTM beads (Molecular Probes, Eugene, OR). These spheres were flowed through the microfluidics
chips using a syringe pump with various flow rates ranging from 0.5mL/hr to 1.5mL/hr. An inverted epifluorescent
microscope (Nikon Diaphot, Nikon, Inc., NY), heavily modified to allow individual filter placement of optical filters was
equipped in this case, with a 475nm band-pass dichroic excitation filter, a 525nm long-pass dichroic beam splitter, and a
525 +/- 12.5nm band-pass dichroic emission filter. This system was used to visualize the beads’ fluorescence properties
as they flowed through the microchannels. Bead concentration used was a 1:1:1 volume ratio of green beads to red
beads to ultrapure water (or 33% green beads, 33% red beads in ultrapure water). Digital movies have been made using a
Nikon 990 Coolpix camera fitted to the microscope, and are being used to provide actual experimental data of inter-
particle spacings to test queuing theory models of single and multi-stage PDMS microfluidic chip design under differing
conditions of particle concentration, and channel widths and geometries. The concept of selectively increasing inter-
particle spacings for "exponentially staging" cell sorting (Leary et al, US Patent pending, 2003) is shown in Figure 6.
Figure 6: (A) Concept of changing inter-particle spacing by using varying channel cross sections and depleting
fluid boluses containing cells not-of-interest, (B) recently constructed Y-channel stage PDMS chip with mixture
of two different colored fluorescent beads showing increases in inter-particle spacing produced by change in
Proc. of SPIE Vol. 6859 68590V-10
60 dg! gI
3.4 Multi-stage microfluidic systems:
Using methods described earlier in this paper we have constructed and are testing two-stage designs to test the
concept that the two sorting stages will scale exponentially rather than arithmetically. Figure 7 shows the CAD design
and the resulting microfluidic PDMS optical biochip resulting from this two-stage design:
3.5 Electronics for microfluidic cytometer/sorter:
We have previously designed and fabricated prototypes of the circuits that will be combined to form the proposed
ASIC (Application Specific Integrated Circuit). We will tailor existing designs to their new application and combine
them as a single module which can be arrayed to process the signal from multiple sorting stages. The new ASIC will
contain circuits for:
Peak track-and-hold (PT&H) – Low-noise operational amplifiers with switched capacitor loads will amplify
APD output and store that on the load capacitor for sampling. Digital logic will control switching will allow
capacitor discharge and tracking of a new sample.
Multiplexing (MUX) – Low-noise operational amplifiers with unity gain for each channel will be combined
sequentially onto a single channel. Channel 1 will connected to the output for one unit of time to acquire sample
1a. Then channel 1 will be disconnected and channel 2 connected to acquire sample 2a. This will be repeated
for every channel before cycling back to channel 1 for sample 1b, and so on. The connecting and disconnecting
happens through digital logic controlled CMOS switches which are integrated with the ASIC. This whole
architecture is called time-division multiplexing, and will be done in the analog domain.
Analog-to-digital conversion (A/D) – Two operational amplifiers configured as comparators will determine
each bit of the 16-bit digital representation of the analog input from the MUX. These stages are combined in a
pipeline using digital logic and precise timing circuitry to give the desired output.
From the preceding description one can see that a small number of functional blocks can be combined to yield a
tremendous degree of functionality. The advantage of this approach is that it allows us to concentrate resources on
developing a small number of high quality building blocks, and then using them in multiple ways. Our recently
published operational amplifiers combine ultra low-power performance with very high sensitivity (Dresher and Irazoqui,
2007). Other operational amplifier designs of ours utilize higher power but introduce chopper-stabilization to increase
sensitivity to further the sampling frequency of interest from the APD. We have designed and fabricated a number of
A/D topologies, including a pipelined A/D, which contains the necessary digital logic and precise timing circuitry for
this work. This design has a 6-bit resolution which we will expand to 16-bits for the cell sorter. All our prior designs
have been fabricated in CMOS using the MOSIS fabrication service (MOSIS, "http://www.mosis.org/", Marina Del Rey,
In this paper we have discussed a number of design issues for engineering small, portable, closed system flow
cytometer/cell sorters. We have discussed alternative designs and the importance of understanding that these systems are
“tightly engineered” integrated devices. Fluidic, optical, and electronic designs greatly affect each other. The primary
consideration is to design as modular a system as possible that allows for realistic throughputs that permit “real”
We would also like to acknowledge the staff of the Birck Nanotechnology Center who provided training and help in the
soft lithography processes.
Figure 7: (A) CAD design
of two-stage branching tree
(B) resulting two-stage
PDMS chip mounted on
glass slides for testing on
an inverted fluorescence
Proc. of SPIE Vol. 6859 68590V-11
6. REFERENCES Download full-text
1. Chabinyc, M.L., Chiu, D.T., McDonald, J.C., Stroock, A.D., Christian, J.F., Karger, A.M., Whitesides, G.M."An
Integrated Fluorescence Detection System in Poly(dimethylsiloxane)
2. Chang, W-J., Akin, D., Sedlek, M., Ladisch, M., Bashir, R., (2003), “Hybrid Poly(dimethylsiloxane) (PDMS)/Silicon
Biochips For Bacterial Culture Applications”, Biomedical Microdevices 5:4, 281-290, 2003.
3. Chediak, J.A., Luo, Z., Seo, J., Cheung, N., Lee, L.P., Sands, T.D. "Heterogeneous integration of CdS filters with GaN
LEDs for fluorescence detection Microsystems" Sensors and Actuators A: Physical, 111: 1-7 (2004).
4. Corio, M.A., Leary, J.F. "System for Flexibly Sorting Particles" U.S.Patents 5,199,576(1993) and 5,550,058 (1996)
and 5,998,212 (1999).
5. Cupp, J.E., Leary, J.F., Cernichiari, E., Wood, J.C.S., Doherty, R.A.:"Rare Event Analysis Methods for Detection of
Fetal Red Blood Cells in Maternal Blood" Cytometry 5: 138-144 (1984).
6. Dresher, R.P., Irazoqui, P.P "A Compact Nanopower Low Output Impedance CMOS Operational Amplifier for
Wireless Intraocular Pressure Recordings," presented at 29th Annual International Conference of the IEEE Engineering
in Medicine and Biology Society, Lyon, France, 2007.
7. Hassell, T.J., Jedlicka, S.S, Rickus, J.L., Irazoqui, P.P. "Constant-Current Adjustale Waveform Microstimulator for an
Implantable Bi-Modal Output Hybrid Neural Prosthesis," presented at 29th Annual International Conference of the IEEE
Engineering in Medicine and Biology Society, Lyon, France, 2007.
8. Hokanson, J.A., Rosenblatt, J.I., Leary, J.F."Some Theoretical and Practical Considerations for MultivariateStatistical
Cell Classification Useful in Autologous Stem Cell Transplantation and Tumor Cell Purging" Cytometry 36: 60-70, 1999.
9. Hu, X., Bessette, P.H., Qian, J., Meinhart, C.D., Daugherty, P.S., Soh, H.T. "Marker-specific sorting of rare cells
using dielectrophoresis. PNAS 102 (44): 15757–15761, 2005.
10. Jackson, W.C., Tran, H.D., O’Brien, M.J., Rabinovitch, E., Lopez, G.P. “Rapid Prototyping of Active Microfluidic
Components Based on Magnetically Modified Elastomeric Materials.” J. Vac. Sci. Technol. 19(2): 596-599 (2001).
11. Leary, J.F. "Rare Event Detection and Sorting of Rare Cells" In: Emerging Tools for Cell Analysis: Advances in
Optical Measurement Technology, Eds. G. Durack and J.P. Robinson, pp. 49-72, 2000.
12. Leary, J.F. "Ultra High Speed Cell Sorting" Cytometry Part A 67A:76–85 (2005).
13. Leary, J.F., Corio, M.A., McLaughlin, S.R: “System for High-Speed Measurement and Sorting of Particles” U.S.
Patents 5,204,884 (1993) and 5,804,143 (1998).
14. Leary, J.F., Frederickson, C.J. “Flow Sorting System and Methods Regarding Same” Full U.S. Patent Application
10/340520 (January 10, 2003). (high-speed microfluidic sorting fluidic architectures)
15. Leary, J.F., He, F., Reece, L.M. "Detection and Isolation of Single Tumor Cells Containing Mutated DNA
Sequences" Proc. Of SPIE 3603: 93-101, 1999.
16. Leary, J.F., Hokanson, J.A., Rosenblatt, J.I., Reece, L.M. “Real-Time Decision-Making for High throughput
Screening Applications” Proc. of SPIE 4260: 219-225, 2001.
17. Leary, J.F., Reece, L.M. Application of Advanced Cytometric and Molecular Technologies to Minimal Residual
Disease Monitoring. Proc. of SPIE 3913: 36-44, 2000.
18. Leary, J.F., Reece, L.M., Hokanson, J.A., Rosenblatt, J.I. “Advanced “Real-time” Classification Methods for Flow
Cytometry Data Analysis and Cell Sorting” Proc. of SPIE 4622: 204-210, 2002.
19. Leary, J.F., Reece, L.M., Yang, X., Gorenstein., D.G. "High-Throughput Flow Cytometric Screening of
Combinatorial Chemistry Bead Libraries for Proteomics and Drug Discovery" Proc. of SPIE 5692: 216 – 223, 2005.
20. Leary, J.F.: "Strategies for Rare Cell Detection and Isolation" In: Methods in Cell Biology: Flow Cytometry (Edited
by Z. Darzynkiewicz, J.P. Robinson, H.A. Crissman), vol. 42: pp. 331-358, 1994.
21. Porta, P.A., Summers, H.D. "Vertical-cavity semiconductor devices for fluorescence spectroscopy in biochips and
microfluidic platforms." J Biomed Opt. 2005 May-Jun;10(3):034001. , 2005.
22. Rosenblatt, J.A., Hokanson, J.A., McLaughlin, S.R., Leary, J.F.: "A Theoretical Basis for Sampling Statistics Appro-
priate for the Detection and Isolation of Rare Cells Using Flow Cytometry and Cell Sorting". Cytometry 26: 1-6; 1997.
23. Szaniszlo, P., Wang, N., Sinha, M., Reece, L.M., Van Hook, J.W., Luxon, B.A., Leary, J.F. "Getting the Right Cells
to the Array: Gene Expression Microarray Analysis of Cell Mixtures and Sorted Cells" Cytometry 59A: 191-202, 2004.
24. Unger, M.A., Chou, H-P, Thorsen, T., Scherer, A., Quake, S.R. "Monolithic Microfabricated Valves and Pumps by
Multilayer Soft Lithography". Science 288: 113-116, 2000.
25. Yang, X., Li, X., Prow, T.W. Reece, L.M., Bassett, S.E., Luxon, B.A., Herzog, N.K., Aronson, J., Shope, R.E., Leary,
J.F., Gorenstein, D.G. “Immunofluorescence Assay and Flow Cytometric Selection of Bead Bound Aptamers” Nucleic
Acids Research 31 (10): 1-8, 2003.
for Microfluidic Applications"
Proc. of SPIE Vol. 6859 68590V-12